The Risk Matrix | AI for Physicians

The signature framework

The AI Use-Case Risk Matrix

Classify any AI workflow in 60 seconds, before you choose a tool. Six tiers, two zones, one habit: know which situation you're in, then move at the speed that situation allows.

Step 1 · Score the workflow

Five factors decide the tier.

The question is never "is this tool safe?" It is "what does this use touch?" Run any workflow through five factors and the tier falls out.

DATA + OUTPUT + CONSEQUENCE + AUTOMATION + OVERSIGHT + CLINICAL PROXIMITY = RISK TIER
DATA

What goes in?

Nothing sensitive, synthetic, de-identified, or PHI. The single biggest driver of tier.

OUTPUT

What comes out?

A private draft for you, teaching material, or something a patient will act on.

CONSEQ

What if it's wrong?

Lost minutes, an awkward slide, or patient harm. Size the failure, not the feature.

AUTO

Who acts on it?

Does a human decision sit between the output and its effect, or does it flow straight through?

OVERSIGHT

Who can catch it?

Is a qualified reviewer in the loop, and would they actually detect the failure mode?

Step 2 · Read the map

Six tiers. Each with a posture, not a prohibition.

Color always carries tier meaning here, never decoration. Green means build now. Amber means verify. Orange and red mean institutional infrastructure, governed rather than feared.

Tier 1

Personal Productivity

  • Planning & triage
  • Nonclinical drafts
  • Study scheduling
Start today
Tier 2

Learning Acceleration

  • Concept explainers
  • Anki-style cards
  • Board questions
Use freely, verify facts
Tier 3

Research & Teaching

  • Evidence tables
  • Journal club
  • Teaching slides
Verify every citation
Tier 4

Clinical-Adjacent

  • Generic handouts
  • Plain-language ed.
  • Synthetic cases
Physician review first
Tier 5

PHI Workflows

  • Chart summaries
  • AI scribes
  • Patient messages
Institutional-grade
Tier 6

Decision Support

  • Diagnosis support
  • Triage / prediction
  • Device-like SW
Governance + validation
Green Zone · Tiers 1 to 4

Build here now. No PHI, no autonomous clinical decisions. Most useful physician AI does not start with patient data.

Institutional · Tiers 5 to 6

Not forbidden. Clinical infrastructure, not personal hacks.

One boundary holds across every tier: no PHI enters an unapproved tool, and no AI output reaches a patient or a chart without physician review. Everything else is a question of workflow design.

Get the one-pager

The Risk Matrix ships free, as the Physician AI Implementation Map.

Join the list and the Map arrives after you confirm your email, alongside the Physician AI Safety Checklist. Module 5 of the course turns it into a six-question screen and scorecard you can run on any workflow.

Get the free starter kit

Start with the free tools

Build better AI judgment before the next tool arrives.

Get the three-resource starter kit, the course preview, and the founding launch details.

Get the free starter kitTry the live demo